\section{Model Context}
\label{sec:context-of-the-model}
\begin{frame}
    \frametitle{(Why) A network?}
    \begin{columns}
        \begin{column}{0.55\textwidth}
            \begin{columns}
                \only<1>{

                    \begin{column}{0.5\textwidth}

                        \begin{figure}[ht]
                            \centering
                            \begin{tikzpicture}[scale=0.65]
                                \input{tikz/plantpollinatornetwork.tex}
                            \end{tikzpicture}
                            \caption{A toy network}
                            \label{fig:plants-pollin}
                        \end{figure}
                    \end{column}
                    \begin{column}{0.3\textwidth}
                        \centering
                        \begin{align*}
                            \begin{pmatrix}
                                1 & 0 & 1 \\
                                1 & 0 & 0 \\
                                1 & 0 & 0 \\
                                1 & 1 & 0
                            \end{pmatrix}
                        \end{align*}
                        \footnotesize
                        Associated bi-adjacency matrix

                    \end{column}
                }
                \only<2>{
                    \begin{column}{0.5\textwidth}
                        \begin{figure}[ht]
                            % \centering
                            \includegraphics[width=1\textwidth]{tikz/applications/baldock/graph-Baldock2019_Bristol.pdf}
                            \caption{Plant-pollinator network from
                                Bristol\newline\cite{baldockSystemsApproachReveals2019}}
                            \label{fig:bristol-network}
                        \end{figure}
                    \end{column}
                    \begin{column}{0.45\textwidth}
                        \centering
                        \begin{figure}
                            \includegraphics[width=0.7\textwidth]{tikz/applications/baldock/mat-Baldock2019_Bristol.pdf}
                            \caption{Adjacency matrix of the network}
                        \end{figure}
                    \end{column}
                }
            \end{columns}
        \end{column}
        \begin{column}{0.4\textwidth}
            \only<1>{
                \begin{itemize}
                    \item Bipartite graph $G = (U,V,E)$
                    \item Encoded in bi-adjacency matrix $Y \in \{0,1\}^{n_1 \times n_2}$
                \end{itemize}}
            \only<2>{
                \begin{itemize}
                    \item Increasingly available
                    \item Ecosystems described by their interactions
                    \item Functional structure for: biodiversity monitoring, robustness, risk of collapse
                \end{itemize}}
        \end{column}
    \end{columns}
\end{frame}

\begin{frame}{Analysis methods for a network}
    Several methods~:
    \begin{itemize}
        \item Metrics at \begin{itemize}
                  \item node level: degree, centrality\dots
                  \item network level: density, nestedness\dots
              \end{itemize}
              \cite{kolaczykStatisticalAnalysisNetwork2009}
        \item \textbf<2>{Node embedding and/or clustering with latent variable models}
              \\\cite{snijdersEstimationPredictionStochastic1997,hoffLatentSpaceApproaches2002}
        \item Node or network embedding with Graph Convolutional Networks
              \\\cite{kipfVariationalGraphAutoEncoders2016a}
    \end{itemize}
\end{frame}

\begin{frame}
    \addtocounter{footnote}{1}
    \frametitle{Bipartite Stochastic Block Model (BiSBM\footnotemark[\thefootnote])}\framesubtitle{\cite{govaertEMAlgorithmBlock2005}}
    \begin{columns}
        \begin{column}{0.40\linewidth}
            \begin{figure}[H]
                \center
                \begin{tikzpicture}[scale=0.35]
                    \input{tikz/lbm.tex}
                \end{tikzpicture}
                \caption{Example of BiSBM}
                \label{fig:LBMvisu}
            \end{figure}
        \end{column}
        \begin{column}{0.51\linewidth}
            \begin{block}{Hierarchical model}
                \vspace{-\baselineskip}
                \begin{align*}
                    \forall q\in[\![ 1, Q_1]\!],\mathbb{P}(Z_i = q) = \pi_q   \\
                    \forall r\in[\![ 1, Q_2]\!],\mathbb{P}(W_j = r) =  \rho_r \\
                    Y_{ij} | Z_i = q, W_j = r \sim \mathcal{B}ern(\alpha_{q,r})
                \end{align*}
                where $|\pi| = Q_1, |\rho| = Q_2, |\alpha| = Q_1 \times Q_2$
            \end{block}
            \begin{block}{Concise BiSBM formula}
                $Y \sim \mathcal{B}ern\text{-BiSBM}_{n_1,n_2}(Q_1, Q_2, \pi, \rho, \alpha)$
            \end{block}
        \end{column}
    \end{columns}

    \footnotetext[\thefootnote]{Commonly Known as \emph{Latent Block Model} (LBM) in the literature.}
\end{frame}

\begin{frame}
    \frametitle{How to compare multiple networks?}
    \begin{figure}[ht]
        \centering
        \begin{subfigure}[ht]{0.475\textwidth}
            \centering
            \includegraphics[width=0.5\textwidth]{tikz/applications/baldock/mat-Baldock2019_Bristol.pdf}
            \caption{Bristol}
        \end{subfigure}
        \begin{subfigure}[ht]{0.475\textwidth}
            \centering
            \includegraphics[width=0.5\textwidth]{tikz/applications/baldock/mat-Baldock2019_Edinburgh.pdf}
            \caption{Edinburgh}
        \end{subfigure}
        \hfill
        \begin{subfigure}[ht]{0.475\textwidth}
            \centering
            \includegraphics[width=0.5\textwidth]{tikz/applications/baldock/mat-Baldock2019_Leeds.pdf}
            \caption{Leeds}
        \end{subfigure}
        \begin{subfigure}[ht]{0.475\textwidth}
            \centering
            \includegraphics[width=0.5\textwidth]{tikz/applications/baldock/mat-Baldock2019_Reading.pdf}
            \caption{Reading}
        \end{subfigure}
        \caption{Adjacency matrices,~\cite{baldockSystemsApproachReveals2019}}
        \label{fig:adj}
    \end{figure}
\end{frame}

\section[Bipartite collection models]{Bipartite network collection models}
\label{sec:extension-of-colsbm-to-bipartite-networks}

\begin{frame}
    \frametitle{First approach: sep-BiSBM}
    \only<1-2>{
        \begin{equation*}
            \forall m \in \{1\dots M\}, Y^m \overset{ind}{\sim} \mathcal{B}ern\text{-BiSBM}_{n_1^m,n_2^m}(Q_1\alert<2->{^m}, Q_2\alert<2->{^m}, \pi\alert<2->{^m}, \rho\alert<2->{^m}, \alpha\alert<2->{^m})
        \end{equation*}}
    \only<3>{
        \begin{figure}[ht]
            \centering
            \begin{subfigure}[ht]{0.42\textwidth}
                \centering
                \includegraphics[width=0.5\textwidth]{tikz/applications/baldock/bisbm-mat-Baldock2019_Bristol.pdf}
                \caption{Bristol, $Q_1 = 3, Q_2 = 3$}
            \end{subfigure}
            \begin{subfigure}[ht]{0.42\textwidth}
                \centering
                \includegraphics[width=0.5\textwidth]{tikz/applications/baldock/bisbm-mat-Baldock2019_Edinburgh.pdf}
                \caption{Edinburgh, $Q_1 = 3, Q_2 = 3$}
            \end{subfigure}
            \hfill
            \begin{subfigure}[ht]{0.42\textwidth}
                \centering
                \includegraphics[width=0.5\textwidth]{tikz/applications/baldock/bisbm-mat-Baldock2019_Leeds.pdf}
                \caption{Leeds, $Q_1 = 3, Q_2 = 2$}
            \end{subfigure}
            \begin{subfigure}[ht]{0.42\textwidth}
                \centering
                \includegraphics[width=0.5\textwidth]{tikz/applications/baldock/bisbm-mat-Baldock2019_Reading.pdf}
                \caption{Reading, $Q_1 = 3, Q_2 = 3$}
            \end{subfigure}
            \caption{Separate BiSBM fit for each network}
            \label{fig:adj-reord}
        \end{figure}
    }
\end{frame}

\begin{frame}
    \frametitle{Several joint models}
    \onslide<1->{    \begin{block}{\emph{iid}-colBiSBM}
            \[
                \forall m \in \{1\dots M\}, Y^m \overset{iid}{\sim}
                \mathcal{B}ern\text{-BiSBM}_{n_1^m,n_2^m}(Q_1, Q_2, \pi, \rho, \alpha)
            \]

            with $\theta = (\pi, \rho, \alpha)$.
        \end{block}}
    \onslide<2>{    \begin{block}{$\pi\rho$-colBiSBM}
            \[
                \forall m \in \{1\dots M\}, Y^m \overset{ind}{\sim}
                \mathcal{B}ern\text{-BiSBM}_{n_1^m,n_2^m}(Q_1, Q_2, \pi\alert{^m}, \rho\alert{^m}, \alpha)
            \]

            with $\theta = ((\pi\alert{^m})_{m=1,\dots, M}, (\rho\alert{^m})_{m=1,\dots,
                        M}, \alpha)$.
        \end{block}
    }
    \begin{itemize}
        \item No shared nodes across networks
        \item Agnostic of structure
    \end{itemize}
\end{frame}
% \begin{frame}
%     \frametitle{Parameter estimation}
%     % DONE say that tau i q m c' is the probability that Zim = q, approximation of the variational probability. Because we impose independence
%     % By maximizing a variational lower bound of the
%     % log-likelihood of the observed data.
%     Maximizing the log-likelihood?
%     \begin{block}{log-likelihood and complete log-likelihood}
%         \[
%             \ell(\mathbf{Y};\theta) = \sum_{\mathbf{Z,W}\in \mathbf{\mathcal{Z}\times\mathcal{W}}} \ell_c(\mathbf{Y}, \mathbf{Z}, \mathbf{W};\theta)
%         \]

%         with $\mathbf{\mathcal{Z}} = \{1,\dots,\alert<2>{Q_1}\}^{\alert<2>{n}},
%             \mathbf{\mathcal{W}} = \{1,\dots,\alert<2>{Q_2}\}^{\alert<2>{n}}$
%     \end{block}
%     \uncover<3>{So, classic algorithm $\Rightarrow$
%         \emph{Expectation-Maximization} (EM).}
% \end{frame}

% \begin{frame}
%     \frametitle{By classic EM}
%     At iteration $(t)$:
%     \begin{itemize}
%         \item[$\bullet$]\textbf{E Step}: calculate
%               $$ \mathcal{Q}(\theta | \theta^{(t-1)}) =   \mathbb E_{\alert<2>{\mathbf Z, \mathbf W | \mathbf Y, \theta^{(t-1)}} } \left[\ell_c(\mathbf Y, \mathbf W, \mathbf Z; \theta)  \right] $$
%         \item[$\bullet$]\textbf{M Step}:
%               $$ \theta^{(t)} = \arg \max_{\theta} \mathcal{Q}(\theta | \theta^{(t-1)})$$
%     \end{itemize}
%     \uncover<2>{
%         \begin{alertblock}{Problem for classic EM}
%             Law of $\mathbf{Z,W|Y},\theta^{(t-1)}$ inaccessible
%         \end{alertblock}}

% \end{frame}
\section{Inference and model selection}
\label{sec:inference-and-model-selection}
\begin{frame}{Parameter estimation}{How ?}
    \begin{align*}
        \ell(\mathbf{Y};\theta) = & \sum_{m=1}^{M} \ell(Y^m;\theta)                                                                                     \\
        =                         & \sum_{m=1}^{M} \log \sum_{\alert<2->{Z^m \in \mathcal{Z}^m,W^m\in\mathcal{W}^m}} \exp\{\ell_c(Y^m,Z^m,W^m;\theta)\} \\
        =                         & \sum_{m=1}^{M} \log\sum_{\alert<2->{Z^m \in \mathcal{Z}^m,W^m\in\mathcal{W}^m}}\exp\{\ell(Y^m | Z^m,W^m;\alpha) +   \\
                                  & \ell(Z^m;\pi) + \ell(W^m;\rho)\}
        % & = \sum_{m=1}^{M} \sum_{i=1}^{n_1^m} \sum_{q=1}^{Q_1} Z_{iq} \log(\pi_q) + \sum_{j=1}^{n_2^m}\sum_{r=1}^{Q_2} W_{jr} \log(\rho_r) \\
        % & + \sum_{i,j}\sum_{q,r} Z_{iq}W_{jr} \log \mathcal{B}ern(Y_{ij};\alpha_{qr})
    \end{align*}

    \onslide<2>{
        EM impracticable since $\mathbf{Z,W|Y}$ intractable due to
        conditional dependency.}
\end{frame}
\begin{frame}{Parameter estimation}{Solution}
    \emph{Variational EM}~\cite{daudinMixtureModelRandom2008,chabert-liddellLearningCommonStructures2024}.
    \begin{block}{Variational approximation of $\mathbf{Z,W|Y}$}
        $\mathcal{R}_{Y^m,\tau}(Z^m, W^m) =
            \mathcal{R}^1_{Y^m,\tau}(Z^m)
            {\color{red}\times}
            \mathcal{R}^2_{Y^m,\tau}(W^m) \Rightarrow$ independence between rows and columns, mean field approximation.
    \end{block}
    \begin{align*}
        \ell (\mathbf{Y};\theta) \geq \color{red}\sum_{m=1}^{M} \bigg(
        \color{black} \mathbb{E}_{\mathcal{R}_{Y^m,\tau}(Z^m,W^m)}
        \left[ \ell_c(Y^m,Z^m,W^m ; \theta) \right] + \\
        \mathcal{H}(\mathcal{R}_{Y^m,\tau}
        (Z^m, W^m))
        \color{red}\bigg) \color{black}
        \eqcolon \mathcal{J}(\mathcal{R}_{\mathbf{Y},\tau};\theta)
    \end{align*}
    where $\theta = (\pi, \rho, \alpha)$ for \emph{iid}-colBiSBM
\end{frame}

\begin{frame}{Selection criterion for $Q_1, Q_2$}
    Integrated Classification Likelihood (ICL)~\cite{biernackiAssessingMixtureModel2000}
    \begin{align*}
        \text{ICL}(\mathbf{Y}, Q_1, Q_2) & = \mathbb{E}_{\mathbf{Z,W|Y}} [\ell_c(\mathbf{Y,Z,W};\hat{\theta})] -\frac{1}{2}\text{pen}(Q_1, Q_2)                     \\
                                         & = \ell(\mathbf{Y};\hat{\theta}) - \mathcal{H}(p(\mathbf{Z,W}|\mathbf{Y};\hat{\theta})) - \frac{1}{2}\text{pen}(Q_1, Q_2)
    \end{align*} For SBM~\cite{daudinMixtureModelRandom2008}.
    \onslide<2->{
        \begin{align*}
            \text{BIC-L}(\mathbf{Y},  Q_1, Q_2) & = \mathbb{E}_{\mathcal{R}_{\mathbf{Y},\hat{\tau}}} [\ell_c(\mathbf{Y,Z,W};\hat{\theta}^{\text{var}})] + \mathcal{H(\mathcal{R}_{\mathbf{Y},\hat{\tau}})} - \frac{1}{2}\text{pen}(Q_1, Q_2) \\
                                                & = \mathcal{J(\mathcal{R}_{\mathbf{Y},\hat{\tau}}, \hat{\theta}^{\text{var}})} - \frac{1}{2}\text{pen}(Q_1, Q_2)
        \end{align*}
    }
\end{frame}
\begin{frame}{Practical problems of choosing $Q_1, Q_2$}
    \begin{alertblock}{Exploration problems}
        \begin{itemize}
            \item Sensitivity to initializations. \uncover<2->{$\rightarrow$ \textbf{Spectral
                          clustering} and \textbf{split \& merge} approach}
            \item Exploration of a 2D grid is costly. \uncover<3->{$\rightarrow$ \textbf{Greedy
                          approach} and \textbf{sliding window}}
        \end{itemize}
    \end{alertblock}
\end{frame}

\section{Application}
\label{sec:application}

\begin{frame}
    \frametitle{Results~\cite{baldockSystemsApproachReveals2019}}
    \only<1>{
        \begin{figure}[ht]
            \centering
            \begin{tikzpicture}[every every node/.style={anchor=south west, inner sep=0pt}, x=1mm, y=1mm]
                \node (struct) at (0,0) {\includegraphics[width=0.8\textwidth]{tikz/applications/baldock/shared-mixture-iid.pdf}};
                \node[isosceles triangle,
                    isosceles triangle apex angle=10,
                    draw,
                    rotate=270,
                    shading = axis,
                    top color=blue!50,
                    bottom color=blue!1!white,
                    anchor=right corner, minimum height=25mm, label={[label distance = 2mm]180:Generalists}, label={[label distance = 2mm]0:Specialists}] (T) at ($(struct.east)+(1.25,8)$) {};

            \end{tikzpicture}
            \caption{Shared structure ($\alpha$ matrix) and proportions ($\pi$ and $\rho$) of the four networks}
            \label{fig:shared-mixture}
        \end{figure}}

    \only<2>{\begin{figure}[ht]
            \centering
            \begin{subfigure}[t]{0.5\textwidth}
                \centering
                \includegraphics[width=0.45\textwidth]{tikz/applications/baldock/colbisbm-mat-Baldock2019_Bristol.pdf}
                \caption{Bristol}
            \end{subfigure}\hfil
            \begin{subfigure}[t]{0.5\textwidth}
                \centering
                \includegraphics[width=0.45\textwidth]{tikz/applications/baldock/colbisbm-mat-Baldock2019_Edinburgh.pdf}
                \caption{Edinburgh}
            \end{subfigure}
            \newline
            \begin{subfigure}[ht]{0.5\textwidth}
                \centering
                \includegraphics[width=0.5\textwidth]{tikz/applications/baldock/colbisbm-mat-Baldock2019_Leeds.pdf}
                \caption{Leeds}
            \end{subfigure}\hfil
            \begin{subfigure}[ht]{0.5\textwidth}
                \centering
                \includegraphics[width=0.5\textwidth]{tikz/applications/baldock/colbisbm-mat-Baldock2019_Reading.pdf}
                \caption{Reading}
            \end{subfigure}
            \caption{\emph{iid}-colBiSBM fit, $Q_1 = 3, Q_2 = 5$}
        \end{figure}}
\end{frame}

\begin{frame}
    \frametitle{Focus on Leeds}
    \captionsetup{font=normalsize}
    \begin{figure}[ht]
        \centering
        \begin{subfigure}[t]{0.5\textwidth}
            \centering
            \includegraphics[width=1\textwidth]{tikz/applications/baldock/bisbm-mat-Baldock2019_Leeds.pdf}
            \caption{sep-BiSBM, $Q_1 = 3, Q_2 = 2$}
        \end{subfigure}\hfill
        \begin{subfigure}[t]{0.5\textwidth}
            \centering
            \includegraphics[width=1\textwidth]{tikz/applications/baldock/colbisbm-mat-Baldock2019_Leeds.pdf}
            \caption{\emph{iid}-colBiSBM, $Q_1 = 3, Q_2 = 5$}
        \end{subfigure}
    \end{figure}
\end{frame}

\begin{frame}{\emph{Bombus}}
    \only<1>{
        \begin{figure}
            \captionsetup{font=normalsize}
            \begin{subfigure}[t]{0.5\textwidth}
                \centering
                \includegraphics[width=0.5\textwidth]{img/baldock/bombus-hortorum.jpeg}
                \caption{\emph{Bombus Hortorum} or garden bumblebee}
            \end{subfigure}\hfill
            \begin{subfigure}[t]{0.5\textwidth}
                \centering
                \includegraphics[width=0.5\textwidth]{img/baldock/bombus-lapidarius.jpeg}
                \caption{\emph{Bombus Lapidarius} or red-tailed bumblebee}
            \end{subfigure}
        \end{figure}
    }
    \only<2->{
        \begin{columns}

            \begin{column}{0.2\textwidth}
                \begin{figure}
                    \onslide<2>{
                        \begin{subfigure}[t]{0.7\textwidth}
                            \centering
                            \includegraphics[width=1\textwidth]{img/baldock/bombus-hortorum.jpeg}
                            \caption{\emph{Bombus Hortorum} or garden bumblebee}
                        \end{subfigure}
                    }
                    \onslide<3>{
                        \begin{subfigure}[t]{0.7\textwidth}
                            \centering
                            \includegraphics[width=1\textwidth]{img/baldock/bombus-lapidarius.jpeg}
                            \caption{\emph{Bombus Lapidarius} or red-tailed bumblebee}
                        \end{subfigure}
                    }
                \end{figure}
            \end{column}
            \begin{column}{0.7\textwidth}
                \begin{figure}
                    \centering
                    \begin{tikzpicture}[every every node/.style={anchor=south west, inner sep=0pt}, x=1mm, y=1mm]
                        \node (struct) at (0,0) {\includegraphics[width=0.8\textwidth]{tikz/applications/baldock/shared-iid.pdf}};
                        \node (gen) at (-29,14.8) {};
                        \node (interm) at (-29,2.5) {};
                        \node (spe) at (-29,-9.8) {};

                        \node[left = 3mm of spe] (@aux) {\phantom{B, E, R}};

                        \node[isosceles triangle,
                            isosceles triangle apex angle=10,
                            draw,
                            rotate=270,
                            shading = axis,
                            top color=blue!50,
                            bottom color=blue!1!white,
                            anchor=right corner, minimum height=42mm, label={[label distance = 2mm]207:Generalists}, label={[label distance = 12mm]357:Specialists}] (T) at ($(struct.north east)+(-1,-2.5)$) {};

                        \only<2>{
                            \node[left = 3mm of gen] (towns_gen_garden) {B, L};
                            \node[left = 3mm of spe] (towns_spe_garden) {\phantom{B, }E, R};
                            \path (towns_gen_garden) edge[->,thick] (gen);
                            \path (towns_spe_garden) edge[->,thick] (spe);
                        }
                        \only<3>{
                            \node[left = 3mm of interm] (towns_interm_red) {L};
                            \node[left = 3mm of spe] (towns_spe_red) {B, E, R};
                            \path (towns_interm_red) edge[->,thick] (interm);
                            \path (towns_spe_red) edge[->,thick] (spe);
                        }
                    \end{tikzpicture}
                    \caption{Shared structure ($\alpha$ matrix) of the four networks}\label{fig:shared}
                \end{figure}
            \end{column}
        \end{columns}
    }

\end{frame}
\section{Conclusion}
\begin{frame}
    \frametitle{Conclusion and perspectives}
    \begin{block}{Summary}
        \begin{itemize}
            \item 4 models including 3 flexible on at least one dimension (adaptability to data).
            \item Jointly detect classic and less classic structures agnostically.
            \item Partition a collection in sub-collections with homogeneous structures.
            \item \texttt{R} package \texttt{colSBM} at \url{https://github.com/GrossSBM/colSBM}
        \end{itemize}
    \end{block}
    \begin{block}{Future work}
        \begin{itemize}
            \item Preprint in redaction
            \item Apply clustering to data from
                  \cite{pichonTellingMutualisticAntagonistic2024,doreRelativeEffectsAnthropogenic2021}.
                  Do interaction type drives the structure of the network?
        \end{itemize}
    \end{block}
\end{frame}